Exponential Modulations

Assoc Prof. Dr. Ho Van Khuong

Tele. Dept., HCMUT
Email: [email protected]

1
Linear versus Exponential Modulations

2
 Exponential Modulation
Consider modulated carrier having a constant amplitude and a time-
varying phase. General representation is:

Here Θc(t) is the total instantaneous angle.

3
 Phase modulation (PM)

Here φΔ is the phase deviation.

Instantaneous angle depends on the modulated signal. The maximum
phase shift is φΔ ≤ 180 (| x(t) | ≤ 1).

Instantaneous frequency:

where

4
 Frequency Modulation (FM)
Instantaneous frequency in frequency modulation is

Here fΔ is the frequency deviation.

The phase is:

It is usually assumed d that φ(t0) = 0, and thus, the waveform of the FM

modulated signal is given by

It has to be assumed that the message has no DC component such that the
integral converges when t→∞. The DC component of x(t) causes a
frequency shift for the carrier and the amount of the shift is fΔ 〈x(t)〉.

5
 Phase and Frequency Modulation (1)
It is difficult to distinguish FM and PM signals from their waveforms. In
both case, the amplitude is constant. The information message resides in
the zero-crossings.

The power of the FM and PM modulated signals is always:

Despite of the similarities, FM modulation works better in noisy conditions.

6
 Phase and Frequency Modulation (2)
It can be seen from the equation of the instantaneous frequency of the FM
modulation that the output signal of the demodulator in the receiver is
proportional to the deviation. It can be increased without changing the
transmission power. Consequently, in FM system the S/N ratio can be
increased independent on the transmission power. However, this
increases the transmission bandwidth.

One idea could be to decrease the bandwidth by using very small

deviation. This does not work in practice because every system has some
noise that would be strongly emphasized.

7
Phase and Frequency Modulation (3)

8
 Narrowband PM and FM (1)
The frequency-domain analysis of the exponential modulation methods
can not be done analytically in general case. Therefore, we consider first
the narrowband case (i.e., narrow signal bandwidth), and then the tone
modulation in narrowband and wideband cases.

Consider a signal that is exponential modulated, by using the phase and

quadrature components:

where

If we assume that φ(t) << 1 rad, then:

9
 Narrowband PM and FM (2)
In this case, the relationship between the spectra of the modulating and
modulated signals is easy to find:

where

It can be seen that the bandwidth of the narrowband FM and PM

modulated signals is 2W (W is the bandwidth of the message signal). The
abbreviations NBFM and NBPM are used for these kind of modulations.

10
 Narrowband PM and FM (3)
Example of Narrowband FM and PM Modulations

The spectra of modulated signals:

The spectra corresponds to the spectra of AM signals. Especially, the

amplitude spectrum of the NBPM-signals is the same as the amplitude
spectrum of the AM signal with the same message signal (and with a
proper modulation index).

11
 Narrowband Tone Modulation (1)
Consider FM and PM modulations when the modulating signal is sinusoidal:

Then

β is the modulation index (it represents the maximum phase deviation and
it is proportional to the tone amplitude). The exponential modulated signal is:

In narrowband tone modulation β << 1 and thus (from Taylor expansion):

12
Narrowband Tone Modulation (2)

13
 Wideband Tone Modulation (1)

It can be shown that:

where Jn(β) are Bessel functions of the first kind:

14
 Wideband Tone Modulation (2)
The modulated signal can be rewritten in the form:

By using the fact J-n(β) = (−1)nJn (β), we end up with

15
 FM Wideband Tone Modulation (1)
Line spectrum of FM with wideband tone modulation:

In principle, the spectrum continues to +/- infinity. It consists of a carrier-

frequency line plus an infinite number of sideband lines at frequencies
ωc± nωm. All lines are equally spaced by the modulating frequency, and
the odd-order lower sidebands are in reverse phase (inverted) relative to
the unmodulated carrier.

16
 FM Wideband Tone Modulation (2)
Phasor Diagram:

The odd-order pairs have a

resultant in quadrature with the
carrier ⇒ this provides the desired
frequency modulation + some
unwanted amplitude modulation.

The even-order pairs have a

resultant collinear with the carrier
⇒ this corrects the amplitude
variations.

17
 Multitone Modulation (1)
Consider:

where

The procedure can be extended to more than 2 tones.

18
 Multitone Modulation (2)
The case f1 << f2, β1 >> β2 ⇒ the double-tone line spectrum is plotted
below:

Note: under the above conditions, each sideband line at ωc ± mω2 looks
like another FM carrier with tone modulation of frequency ω1.

19
Periodic Modulation: Pulse-Train FM Modulation
In the figure:

(See Example 5.1-3, p. 198, [1])

20
Bessel Functions (1)

21
Bessel Functions (2)

22
 Bessel Functions (3)
Recall that the wideband tone modulated signal can be written as:

The carrier J0(β) term varies with modulation index, and it also carries the
message information. For some value of β it is equal to zero.

The number of essential sideband lines depends on β.

The bandwidth is wide for large values of β.

The spectrum decreases monotonically when n/β > 1, especially when

β >> 1.

23
FM Signal Bandwidth–Wideband Tone Modulation (1)
In principle, FM signal has an infinite wide spectrum. In practice, the
bandwidth has to be limited. This causes nonlinear distortion in detected
signal. The amount of distortion is proportional to the used bandwidth.

In the following, some criteria are given for the needed bandwidth, in such
a way that the distortion would not be significant.

Because the analysis can not be done for general case, a sinusoidal is used
as a modulating signal.

The modulating signal is thus:

The essential question is therefore: What are the significant spectral lines
around the carrier? These lines are at frequencies:

24
FM Signal Bandwidth–Wideband Tone Modulation (2)
It was noted that the values of Jn(β) decrease fast when |n| > β, especially
when β >> 1. Thus the significant spectral lines are determined from the
condition:

Consequently, the needed frequency range is fc ± βfm = fc ± AmfΔ. Usually,

only few components are needed outside this range.

For all the cases (i.e., even if β ≤ 1), the closest spectral spikes to the
carrier should be taken into consideration, that is the case where n = ± 1.

25
FM Signal Bandwidth–Wideband Tone Modulation (3)
Example: Tone-modulated line spectra (significant line spectra only). (a)
FM or PM with fm fixed, (b) FM with AmfΔ fixed.

26
FM Signal Bandwidth–Wideband Tone Modulation (4)
Let’s define that the spectral lines for which |Jn(β)| > ε are significant (i.e.,
the number of significant sidebands).

In practical applications 0.01 < ε < 0.1.

Let’s define the function M(β) > 1 in such a way that JM(β) ≥ ε and
JM+1(β) < ε .

The bandwidth is thus:

Based on practical experiments:

ε = 0.1 is not often tight enough
ε = 0.01 is sufficient for all the cases, but is usually too tight.
When β ≥ 2, the chosen M(β) = β+2 is within these limits and has turned
out to be a good criterion.

27
FM Signal Bandwidth–Wideband Tone Modulation (5)

28
FM Signal Bandwidth–Wideband Tone Modulation (6)
Maximum Bandwidth in Tone Modulation:
For tone modulation, the worst-case situation with respect to the
bandwidth is the case where the frequency and amplitude have the
maximum values (maximum-amplitude maximum-frequency tone):

In this case, the bandwidth is (assuming that β ≥ 2):

Note: This case does not correspond to the maximum modulation index β.
Any other tone having fm < W and/or Am < 1 will require less bandwidth
even though β may be larger.

29
 FM Signal Bandwidth (7)
Maximum Bandwidth of FM Signal:
In the general case, the bandwidth of arbitrary modulating signals
(amplitude ≤ 1, bandwidth W) is usually not greater than the above
mentioned maximum bandwidth of the tone modulation. Therefore, this can
be considered as a criterion for determining the bandwidth of an FM signal.

Then, we deal with deviation ratio D instead of modulation index β:

And the bandwidth can be written as a function of the deviation:

30
 FM Signal Bandwidth (8)
Different approximations are used for function M(D) (which is the same
as M(β) above):

 Often D is in the range 2 < D < 10 and the approximation above

M(D) = D + 2 gives the best result. Then

 Carson's rule is also often mentioned:

31
 FM Signal Bandwidth (9)
Example: Commercial FM Bandwidth

Required bandwidth: BT = 2 (5 + 2) × 15 kHz = 210 kHz (Carson's rule:

180 kHz).

In FM radios: bandwidth of at least 200 kHz required.

Tone modulation:

32
 PM Signal Bandwidth
The above rules also apply to PM modulation if we replace the deviation
D with the maximum phase deviation φΔ. Necessary bandwidth for PM
modulation is then:

This is different than for FM modulation in the sense that φΔ is

independent of the bandwidth W.

Note: when the signal has discontinuities (e.g., a rectangular pulse

train), the above estimates of the necessary bandwidth become invalid.

33
34
 Generation and Detection of FM
Next we will consider:
 the effect of linear filtering for FM signal
 the effect of nonlinear elements for FM signal
 FM modulation methods
 FM detection
Here we mainly consider FM modulation, but most of the results can be
also used for PM modulation.
The constant amplitude property of PM and FM modulation is a great
advantage compared to the linear modulation methods.
We will see that nonlinear elements do not cause problems for
exponential modulations. Therefore, nonlinear amplifiers can be used in
the transmitter, which have much better efficiency than linear amplifiers.
In addition, the peak power of FM transmitters is equal to the average
power.

35
Distortion of FM Signals Caused by Linear Filtering (1)

This is very difficult to analyze in closed form and in general case. We

consider the case where the amplitude and phase responses depend
linearly on the frequency ⇒ linear distortion. Then, one can show that:

36
Distortion of FM Signals Caused by Linear Filtering (2)
It can be seen that in the envelope of FM signal, there is an additional
message-dependent AM component with modulation index μ = K1 fΔ/fc.
Here t0 is the carrier delay and t1 is the envelope delay.

As it will be seen, AM component can be attenuated before the detection.

However, it is assumed that the attenuation is not so high in the frequency
band of the FM signal, so that it won't decrease the S/N ratio.

The delay terms t0 and t1 do not cause any problems. Nonlinear phase
response, however, would be a problem and it would cause distortion to
the detected signal.

37
Distortion of FM Signals Caused by Linear Filtering (3)
Hard Limiter:
Due to filtering ⇒ AM components. Therefore, we need to remove these
AM components ⇒ The limiter (or clipper) is used to remove AM
components before detection:

38
Distortion of FM Signals Caused by Linear Filtering (4)
Effect of Nonlinearities, Limiter and Frequency Multiplier:
The limiter and other memoryless nonlinear elements cause harmonics in
the frequencies ± nfc. In the general case, the output signal of exponential
modulation is:

Extra harmonics can be attenuated by filtering.

Amplitude limiter preserves the fundamental frequency:

The frequency multiplier preserves some of the harmonics:

Then the frequency and phase deviation increases n times higher. This is
utilized in some FM modulators.

39
 FM Modulators (1)
Such kind of circuits are needed where the phase or the frequency of the
modulated signal depends linearly on the modulating signal.

1. Direct FM Modulator
Conceptually, a direct FM requires only a Voltage Controlled Oscillator
(VCO) whose oscillation frequency has a linear dependence on the
applied (input) voltage.

The implementation can be based on

 tube ( fc ≥ 1 GHz)
 semiconductors (varactors)

40
 FM Modulators (2)
VCO Circuit with Varactor Diode:

Advantages:
 a large deviation can be implemented directly
 simple implementation
Disadvantages:
 it is difficult to stabilize the carrier frequency (carrier frequency
tends to drift)

41
FM Modulators (3)

42
 FM Modulators (4)
2. Indirect FM Modulator

In practice, the frequency multiplier is implemented in cascade where

each stage has a factor of 2 or 3.

43
 FM Modulators (5)
Example: FM voice radio

44
 Phase Modulators (1)
Although we seldom transmit a PM wave, we are still interested in phase
modulators because:
 the implementation is relatively easy
 the carrier can be supplied by a stable frequency source, such as a
crystal-controlled oscillator
 PM modulators can be used to generate FM modulators:
integrating the input signal to a phase modulator produces a
frequency modulated output.

45
 Phase Modulators (2)
 Narrowband Phase Modulation
NBPM is easy to implement. It is needed as a part of FM modulator.
(a) Implementation based on the narrowband approximation

Assumption:

46
 Phase Modulators (3)
(b) Switching-circuit NBPM modulator: (for large phase shifts)

47
 FM Detection (1)
The task of a frequency detector (i.e. discriminator) is to produce an
output voltage that is linearly dependent on the instantaneous frequency.
Frequency detector categories are:
 FM-AM-conversion
 Phase-shift discriminator (quadrature detector)
 Zero-crossing detection
 Phase locked loop (PLL)

Note: Analog PM detection is not discussed here because it is seldom

needed in practice and, if needed, can be accomplished by integrating the
output of a FM detector.

48
 FM Detection (2)
 FM - AM conversion
FM-AM conversion is produced by a transfer function having
magnitude (amplitude) distortion, as, for example, the time
derivative

49
FM Detection (3)

50
 FM Detection (4)
The limiter is needed to remove any spurious amplitude variation from the
received signal.

In practice, the derivative can be approximately obtained by using a

tuned circuit.

Extended linearity can be achieved by a balanced discriminator. A

balanced discriminator consists of two tuned circuits, one below and one
above the carrier frequency. This results in a wider linear range in the
amplitude curve.

51
 FM Detection (5)
 FM Detection Based on Zero-Crossings

. After DC block:

52
 Interference (1)
Interference is caused e.g., by signals from other stations or cross-talk
between cables. So there might be two (or more) signals from a human
source. Interference can be categorized as follows
 co-channel interference (same carrier frequency)
 adjacent channel interference (neighbour carrier frequency)

Lets consider the case, where the desired signal is unmodulated carrier:

53
 Interference (2)
 Additive Interference in Unmodulated Carrier

If ρ << 1 then

54
 Interference (3)
If ρ << 1 then

If ρ >> 1 then

Put v(t) in the form:

For ρ << 1, interference can cause tone modulation at frequency fi

with amplitude modulation (modulation index µ = ρ) and frequency
modulation or phase modulation (modulation index β = ρ).
For ρ >> 1, interference still cause tone modulation at frequency fi
with amplitude modulation (modulation index µ = ρ-1) but the phase
corresponds to a shifted carrier frequency fc + fi plus the constant φi.
55
 Interference (4)
 Additive Interference for the Demodulated Signal
For different modulation methods, the detected signal is (here φi = 0 and
fi ≤ W, ρ << 1):

 In AM and PM a tone interference produces a tone to reception

whose amplitude is comparable to ρ.
 In FM, the interference is more severe the more distant the
interfering tone is from the carrier (adjacent channel interference).
 On the other hand, it can be seen that FM is much better that AM or
PM when the disturbing frequency is close to the carrier frequency
(co-channel interference).

56
 FM Capture Effect
Capture effect is a phenomenon that takes place in FM systems, when
there are two FM modulated signals in the same channel and the
amplitude of these signals is nearly the same.

The stronger dominates or “captures” the output (annoying results

when listening to a distant FM station). When the strengths of the two FM
signals begin to be nearly the same, the capture effect may cause the
signals to alternate in their domination of the frequency. Small variations
in the relative amplitude levels cause sudden change in the station.

57
 De-emphasis and Pre-emphasis (1)
In FM detection, there is usually more interference in the higher
frequency than lower frequency (this applies also to noise). These high
frequencies of the interferer can at times be larger in amplitude than the
high-frequency content of the modulating signal.

This can be compensated by using de-emphasis (e.g., a LPF) after the

detection. De-emphasis has to be compensated by pre-emphasis before
FM modulation.

58
 De-emphasis and Pre-emphasis (2)
The goal is that when |f | ≤ W. Usually 1st degree
transfer functions are used:

Pre-emphasizer filter acts as a differentiator at high frequencies (HPF),

the output spectrum being proportional with fX(f) for | f | >> B .
Differentiating a signal before FM is equivalent with PM modulation ⇒
pre-emphasized FM is actually a combination of PM and FM, combining
the advantages of both with respect to interference.

59
De-emphasis and Pre-emphasis (3)

60
De-emphasis and Pre-emphasis (4)

61
 Noise in CW Modulation
In this part of chapter 3, we will discuss about the effects of the channel
noise for the linear and exponential modulations.

Noise analysis is an essential part of developing different communication

systems. The goal is to develop such a system where the effects of noise
are minimized.

If the effect of channel noise can be decreased in the receiver end, then we
can use lower transmission power. This is especially important, e.g., in
mobile phones and in satellite communications, where the power of the
transmitter should be minimized.

62
 Noise in Linear Modulation (1)
 The model of an Analog Transmission System

Assumptions:
 Message signal x(t) is ergodic with bandwidth W,

 The attenuation of the channel is L and the channel is distortionless (or

equalized).
 Received power:
 Transmitted waveform:

63
 Noise in Linear Modulation (2)
The part of the receiver before the detection is modelled by Band Pass
Filter (BPF) having unit gain and bandwidth of BT. For instance, in super-
heterodyne receiver this is the bandwidth of the IF amplifier. The excess
of amplification and mixing are the same for the signal and noise, and
thus, it is not necessary to take them into account.

The input signal for the detector (pre-detection signal) can be given by:

where n(t) is noise.

In the ideal case, the demodulated signal for different types of

detectors can be expressed as follows:

64
 Noise in Linear Modulation (3)

The above formula assumes perfect carrier synchronization in

synchronous detection.

The term corresponds to the removal of the DC offset in the envelope

detector. This can be included in other detectors as well.

If the detector constant were not equal to unity, it would have the same
effect on the signal and noise.

The last part of the system after the detector is a Low Pass Filter HD(f).
The de-emphasis can be included in this filter.

65
 Noise in Linear Modulation (4)
 The Input Noise Power for the Detector
The mean square of the noise at the input of the detector is:

Here NR is the power of the noise at the input of the detector. It

includes the channel noise and the noise from the receiver front end. The
combination of these noises is assumed to be white, having the power
spectrum of G(f) = N0 /2. Therefore, the power spectral density (PSD) of
the noise at the input of the detector after filtering with HR(f) is given by:

Recall: the noise power is the integral of the PSD over the bandwidth of
interest.

66
 Noise in Linear Modulation (5)
Let’s assume that the filter is almost an ideal Band Pass Filter having the
bandwidth of BT, and then the noise power at the input of the detector is:

67
 Noise in Linear Modulation (6)
 S/N Ratio Before the Detection

S/N ratio before the detection is defined as:

This can be rewritten as:

where

This is actually an upper bound for S/N ratio. Nonlinearities in

practical system decrease this ratio.

Note that γ corresponds to the maximum S/N ratio in analog baseband

transmission with the same values of SR and N0. S/N ratio before
detection is never higher than γ.

68
 Noise in Linear Modulation (7)
 Bandpass Noise
Let n(t) be a stationary Gaussian noise. It is assumed that it does not
have DC component and thus

It can be expressed by using the in-phase and quadrature components:

where ni(t) and nq(t) are also stationary Gaussian noise, they are
independent and have the properties:

69
 Noise in Linear Modulation (8)
 Build-up of bandpass noise components

The power spectral densities of quadrature components are:

70
 Noise in Linear Modulation (9)
Bandpass noise for some modulation formats: (a) General case (e.g. VSB);
(b) FM, PM, DSB, AM (α = 1/2); (c) SSB (USSB or LSSB, α = 0 or 1)

71
 Noise in Linear Modulation (10)
 The Envelope and Phase of Noise
The bandpass noise can be also expressed in the envelope - phase form:

The following relations hold between the above form and the quadrature
form:

The envelope An has the Rayleigh distribution:

with mean and mean square are given by

72
 Noise in Linear Modulation (11)
The probability that An exceeds value a is then (see [1] for proof):

The phase φn has a uniform distribution over [0, 2π].

Hence

(Use:

73
 Noise in Linear Modulation (12)
 Linear Modulation with Noise
The model for the receiver:

S/N ratio at the input of the detector:

Signal and noise at the input of the detector:

The problem is now to determine the output signal yD(t) and the S/N ratio
at the output denoted by (S/N)D for different modulations and detector
types. (S/N)D: post-detection SNR.

74
 Noise in Linear Modulation (13)
 Synchronous Detection for DSB
Ideal synchronous detection separates the in-phase component of v(t).
For example, in DSB modulation:

If the post-detection filter (LPF HD(f)) approximates an ideal LPF with

bandwidth W, then

We assume that the frequency response of the filter before the detector
is nearly rectangular, having the bandwidth of BT = 2W centred at fc.
Hence, the power spectrum of the noise is:

This is lowpass-filtered white noise.

75
 Noise in Linear Modulation (14)
Because

the S/N ratio after the detection (post-detection SNR) is

Comment: post-detection SNR = 2× pre-detection SNR (for DSB).

For DSB, BT = 2W, thus the post-detection SNR is equal to the

baseband SNR:

Therefore, DSB modulation with ideal synchronous detection has the

same performance as the baseband transmission if the noise power
spectrum is constant.

76
 Noise in Linear Modulation (15)
 Synchronous Detection for AM
By assuming that μ =1, AM signal is of the form:

If the synchronous detector removes the DC component, yD(t) has the

same form as for DSB:

Now,

Then, the S/N ratio after the detection is

Because , then the post-detection SNR is at

most equal with half of the baseband SNR.

77
 Noise in Linear Modulation (16)
 Synchronous Detection for SSB and VSB
For SSB (as well as for VSB with a small vestige):

Synchronous detection rejects the quadrature component of both signal

and noise, leaving:

In this case,

The S/N ratio after the detection is:

Thus, SSB has the same post-detection S/N ratio as DSB.

78
 Noise in Linear Modulation (17)
For VSB+C, the S/N ratio after the detection is approximately the
same as for AM:

assuming that BT ≈ W and µ ≈ 1.

79
 Noise in Linear Modulation (18)
 Properties of Synchronous Detection
By assuming that the average transmission power is fixed, the results
derived above show that:
 The message and noise are additive at the output if they are additive at
the detector input.
 If the pre-detection noise spectrum is reasonably flat over
transmission bandwidth, the destination noise spectrum is essentially
constant over the message bandwidth.
 Relative to S/N ratio after the detection, VSB and SSB have no
particular advantage over AM and DSB, respectively.
 The same S/N ratio can be achieved with the linear modulations
as with the baseband transmission (assuming flat noise spectrum).
 Modulations with suppressed carrier (DSB, SSB) provide better S/N
ratio than modulations where the carrier is not suppressed (AM,
VSB+C).

80
 Noise in Linear Modulation (19)
Note: If the comparison is based on the peak envelope power (instead of
average power) and the modulating signal is reasonably continuous, SSB
has about 3 dB better S/N ratio than DSB and about 9 dB better than AM.

If modulating signal contains discontinuities, SSB is worse than DSB.

81
 Noise in Linear Modulation (20)
 Envelope Detection
Next we will consider envelope detection, which is usually used for
detection of AM. The signal for the detector input can be expressed as:

where we are still taking µ = 1.

From phasor diagram, we obtain:

82
 Noise in Linear Modulation (21)
This expression is further developed in two different cases: (1) the level of
the signal is large compared to the noise, (2) the level of the signal is small
compared to the noise.

(1) Large signal power:

In this case, the following approximation is done:

The ideal envelope detector gives:

which is the same result as in the synchronous detection.

(2) Small signal power:

In this case, the noise is written in the form:

83
 Noise in Linear Modulation (22)
From the phasor diagram, we obtain:

Therefore,

where
Through signal and noise were additive at input, the detected
message term is multiplied by noise in form of cosφn(t), which
is random. Thus, the message is hopelessly mutilated.

84
 Noise in Linear Modulation (23)
There is a threshold effect when using the envelope detection: for
moderate-to-high (S/N)R ratio (i.e., pre-detection SNR), the envelope
detector works as well as the synchronous detector. But, under certain
threshold SNR level, the message signal is lost in noise.

It is difficult to determine the threshold. One criterion is the level where

Ac ≥ An with the probability of 0.99. Then (S / N)R = 4ln10 ≈ 10 dB.

In audio broadcasting, the smallest useful S/N ratio is about 30 dB, and
thus, the threshold effect is not a problem. However, the threshold effect
is a problem in some digital modulation methods, which are used in much
lower S/N ratios.

Note: synchronous detection does not suffer of threshold effect.

85
 Noise in Exponential Modulation (1)
The inherent nonlinear nature of exponential modulations (PM, FM) leads
to analytic difficulties. Therefore, we begin with the large signal-to-noise
ratio condition.

Received signal:

In either case, the carrier amplitude remains constant, so

Pre-detection SNR

This is often called the carrier-to-noise ratio (CNR). The pre-detection

BPF is assumed to have a nearly ideal response with bandwidth BT.

86
 Noise in Exponential Modulation (2)
Model for detection of exponential modulation with noise:

Before detection, the limiter suppresses any amplitude variation

represented by Av(t). Input signal to detector is:

87
 Noise in Exponential Modulation (3)
From the phasor diagram:

It shows that

88
 Noise in Exponential Modulation (4)
 Simplified noise model: Large signal powers:
Consider the case where Ac >> An(t) so (S/N)R >> 1. We can use the
small argument approximation to the arctangent function. Next we
assume that φn(t) - φ(t) = φn(t). We assume this step by recalling that φn(t)
has a uniform distribution over [0, 2π]; hence, in the sense of ensemble
averages, φn(t) - φ(t) differs from only by a shift of the mean value. With
these assumptions we obtain:

where is the quadrature component of noise. In this

model, signal phase and equivalent phase noise are additive:

89
 Noise in Exponential Modulation (5)
 Consider the output of a phase detector, when φ(t) = 0. The PM
post-detection noise spectrum is:

If the out-of-band noise is removed by an ideal low-pass filter with

bandwidth W, then the output of noise power will be

90
 Noise in Exponential Modulation (6)
 Next, consider the output of a frequency detector with φv(t) = ψ(t).
The output of the instantaneous frequency noise after frequency
discriminator is by definition:

It can be proved that the FM post-detection noise power spectrum

is then

91
 Noise in Exponential Modulation (7)
If we again take the post-detection filter to be an ideal LPF, the
destination noise power will be:

If we incorporate de-emphasis filtering (i.e. a low pass filter

connected after the detector in order to reduce more the noise)
such that

then the noise power will be

92
Noise in Exponential Modulation (8)
In the case of W/Bde >> 1, then

since

Comments:
– Post-detection noise spectral densities have out-of-band
components that must be removed by LPF.
– PM-noise is flat whereas FM-noise spectrum increases
parabolically (higher baseband frequencies suffer from
more noise contamination than lower frequencies). De-
emphasis/pre-emphasis filtering compensates for this
effect.
– The destination noise power decreases when signal
power SR increases (noise quieting).

93
 Noise in Exponential Modulation (9)
Destination (or Post-detection) SNR for PM:
Now for the case: φ(t) ≠ 0 (still at high signal level). The demodulated
signal plus noise in PM system with φ(t) = φ∆x(t) is

The post-detection filter passes the signal term φ∆x(t), so

From previous result for ND, then we obtain:

PM gives an improvement over DSB-modulation of exactly φ∆2Sx.

But because of φ∆ ≤ π, the improvement is no greater than π2, or about
10 dB.

94
 Noise in Exponential Modulation (10)
Destination (or Post-detection) SNR for FM:
Also for the case: φ(t) ≠ 0 (still at high signal level). The demodulated
signal plus noise in FM system with is

The post-detection filter passes the signal term f∆x(t), so

From previous result for ND, then we obtain:

For deviation ratio D = f∆ /W, we rewrite:

95
 Noise in Exponential Modulation (11)
It seems that the destination SNR for FM system can be made
arbitrary large by increasing D.

But there are some limitations. Recall that the deviation ratio increases
when the transmission bandwidth BT increases.

Therefore, exchage increased bandwidth for reduced transmitter

power, while keeping the destination SNR constant.

In the case of wideband FM (WBFM), , so

FM has better SNR than DSB when:

96
 Noise in Exponential Modulation (12)
Destination SNR for FM using de-emphasis filtering (Bde << W):

The de-emphasis improvement factor is

97
 Noise in Exponential Modulation (13)
Example: FM-Broadcastings

 without pre-emphasis:

 with pre-emphasis:

98
 Noise in Exponential Modulation (14)
Therefore if DSB or SSB system could be exchanged to FM system, 640
fold transmission power savings could be achieved. Note, however that
the required transmission bandwidth is now about 220 kHz /15 kHz = 15
times larger! Also, a problem is the FM threshold effect that we discuss
in the next slides.

99
 Noise in Exponential Modulation (15)
 FM-Threshold Effect
Until now: large signal level. Now, let’s focus on small signal condition:
Ac << An(t). Then the phase at detector input is:

Therefore in the case (S/N)R <<1, the noise dominates and the message
contained in φ(t) can not be recovered.

In the case of (S/N)R ≈1:

100
 Noise in Exponential Modulation (16)
Even small noise variations may then produce large spikes to the
demodulated FM signal.

In the case of tone modulation, the total output noise becomes:

The second term is the contribution of the spikes.

The above ND equation is described in the Figure in next slide with two
values of deviation ratio D. The sudden drop-off of these curves is the FM
threshold effect. Below threshold, noise captures the output signal (also
called noise multilation).

101
 Noise in Exponential Modulation (17)
FM noise performance without pre-emphasis:

102
 Noise in Exponential Modulation (18)
Therefore, when the system is operating near the threshold level, small
variations of signal powers may cause significant changes in the output
signal.

Experimental studies show that typical threshold level for pre-detection

SNR is (S/N)R ≈ 10 dB. Therefore, we define the threshold point as:

The corresponding threshold for baseband SNR is therefore:

Since . Here, we used BT ≈ 2(D+2)W.

Note: The threshold point and threshold for baseband SNR also apply to
PM with D replaced by φ∆.

103
 Noise in Exponential Modulation (19)
Example of threshold point for FM:
FM performance above threshold
is impressive (improved when
increasing D, FM performance is
better than baseband transmission
having (S/N)D = γ).
Now, considering γ fixed (e.g. 20 dB).
When D = 2 (BT ≈ 7W), we are above
threshold and (S/N)D = 28 dB. But,
when increasing D (e.g. D = 5, then
BT ≈ 14W), we are below threshold,
and the signal can not be recovered
due to noise multilation. The system
performance may deteriorate with
increased deviation.

104
Comparison of CW-Modulation Systems (1)

105
 Comparison of CW-Modulation Systems (2)
 Suppressed carrier methods are superior to conventional AM due to: better
SNRs and no threshold effect.

 When bandwidth conservation is important ⇒ SSB and VSB are the best.

 However, efficient linear modulators/demodulators are difficult to

implement.

 Synchronous detection is much more complex than envelope detection.

However, there is no threshold effect for synchronous detection. Threshold
effect in linear modulations is associated with envelope detection.

 Exponential CW modulation can increase the post-detection SNR

substantially, with only moderate complex implementation.

 The signals with significant low-frequency components are best modulated

with DSB or VSB modulation methods.

106
 Comparison of CW-Modulation Systems (3)
 CW modulation systems including 12dB pre-emphasis improvement for
FM (performance below threshold
is omitted):

• For PM: b is limited to 10

(due to phase deviation
limitation)
• For FM: b is unlimited
(as long as the system
above threshold)
• Note: For equal b, FM is
superior to PM.

107
 Comparison of CW-Modulation Systems (4)
 FM/PM signals require more bandwidth than linear modulation methods.

 In FM/PM cases it is possible to use efficient amplifiers (e.g. class C),

while for AM/DSB/SSB we need highly linear amplifiers (classes A or
AB), which are more expensive.

108
 Chapter 4:

Sampling and Pulse Modulation

109
 Sampling Theory and Pulse Modulation
Mathematical functions and electric signals are frequently displayed as
continuous curves. A smooth curve drawn can be displayed by using
samples that have sufficiently close spacing. When the samples are
represented as, e.g., voltage pulses, we obtain a discrete-time signal.

In information transmission, we can use samples instead of a continuous

time signal. Instead of CW-modulation methods, we can use pulse
modulation methods.

A digital signal is obtained by representing the discrete samples as

discrete number of amplitude values (quantization), usually by using the
binary system. Number of bits determines the number of discrete values.
In digital transmission systems are used digital modulation methods.

Next we consider the sampling and the properties of discrete-time signals.

110
 Ideal Sampling (1)
The act of sampling an analog signal at precise time intervals is simply
multiplying the input signal by sampling pulses.

In the ideal sampling, pulse train consists of impulses (or square pulses
when τ → 0):

The ideal sampling function:

Discrete-time signal (ideal sampled wave) is then:

since

111
 Ideal Sampling (2)
The spectrum of the discrete-time signal is:

The spectrum is periodic: Sampling replicates spectrum of continuous-

time signal at integer multiples of sampling frequency.

112
 Sampling Theorem (Shannon Theorem)
Shannon Theorem: A bandlimited continuous-time signal, whose
bandwidth is W, can be exactly reconstructed from its samples, when the
sampling frequency fs ≥ 2W.

Critical sampling frequency fs = 2W is called Nyquist frequency.

If fs < 2W, components of the periodical spectrum fall within the message
band (aliasing).

113
 Ideal Reconstruction (1)
For reconstruction, an ideal low-pass filter with bandwidth B can be used
with frequency response given by

Requirement for the bandwidth B:

(an exact value of B is not critical as long as the above conditions hold)

Reconstructed signal (after low-pass filtering):

The impulse response of the ideal reconstruction filter is

114
 Ideal Reconstruction (2)
Reconstructed signal can be represented as

When B = fs /2, K = 1/fs, and td = 0 then

The correct values are reconstructed at the sampling instants t = kTs,

because the sinc function is zero at all the times excepting one. Between
sampling instants, x(t) is interpolated using sinc function. This is the
reason for which an LPF is often called an interpolation filter.

115
Ideal Reconstruction (3)

116
 Practical Sampling (1)
Practical sampling differs from ideal sampling in 3 main aspects:
(1) The sampled wave consists of pulses having finite amplitude and
duration rather than Dirac impulses.
(2) Practical reconstruction filter are not ideal filters
- Passband should be designed to be good enough.
- The lack of stopband attenuation may cause spurious frequency
components.
- The filter complexity depends on the width of transition band.

117
 Practical Sampling (2)
(3) Messages are not strictly bandlimited ⇒ aliasing. (It is the
consequence of signals that are time-limited)

The aliasing effect is far more serious than spurious frequencies passed by
non-ideal reconstruction filters.

To combat aliasing we need to filter the message as much as possible

before sampling (to reduce its bandwidth).

Sampling is usually done at a rate slightly greater than the nominal

Nyquist rate (i.e., oversampling).

118
 Practical Sampling (3)
Example: Voice signal

Spectrum extends well beyond 10 kHz.

Most of the energy is concentrated in the range 100-600Hz.

A bandwidth of 3 kHz is sufficient for intelligibility.

Typical bandwidth for telephone systems is 3.4 kHz and sampling

frequency is 8 kHz (higher than Nyquist frequency). To avoid the
disturbing aliasing effect, 30 ÷ 40dB attenuation is needed for higher
frequencies.

In HIFI audio systems (e.g. CD-Player), sampling frequencies are 40 ÷ 50

kHz.

119
 Analog Pulse Modulation (1)
PAM: Pulse Amplitude Modulation
PDM: Pulse Duration Modulation
PPM: Pulse Position Modulation

If message waveform is adequately described by periodic sample values, it

can be transmitted using analog pulse modulation.

The sample values modulate the amplitude of a pulse train (PAM), its
width (PDM) or its position (PPM).

120
 Analog Pulse Modulation (2)
 Pulse Modulation:
Used for both analog and digital signals (here emphasis on the analog
pulse modulation).
Analog signals must first be converted to digital signals, which
involves “sampling.”. Three main steps:
 The first step is low-pass filtering of the analog signal (to
remove/diminish the high frequency components in order to
avoid/diminish aliasing).
 The second step is sampling the analog signal at the Nyquist rate
(at least twice the maximum frequency component in the
waveform).
 The third step is transforming the pulses into a digital signal
(quantization).

121
 Analog Pulse Modulation (3)
 Pulse Amplitude Modulation (PAM)
Obtained by sample-and-hold techniques

122
 Analog Pulse Modulation (4)
Unipolar PAM signal is defined by

The PAM spectrum:

With pulse duration τ, we assume that

Then, for adequate pulse resolution, it requires:

The bandwidth of the PAM is considerably greater than the message

bandwidth. PAM is used mainly as a first stage in digital systems.

123
 Analog Pulse Modulation (5)
 Pulse-Duration and Pulse Position Modulation
Constant amplitude, pulse duration (width) or position varies in direct
proportion to the sample values of x(t).
 PDM (or PWM): At each sampling instant a pulse is generated
with a fixed amplitude and a width which is proportional to the
sample values of the modulating wave.
 PPM: At each sampling instant a pulse is generated with a fixed
amplitude and width. Each pulse is delayed depending on the
relative amplitude of the modulating signal.

124
Analog Pulse Modulation (6)

125
 Analog Pulse Modulation (7)
One of the big advantages of pulse modulation is the possibility to use
time division multiplexing, because pulses used are of a relatively short
duration then sufficient space remains between the samples for insertion
of pulses from other sampled signals.

The disadvantages of pulse modulation techniques are that:

- the pulses require a wide bandwidth to be transmitted.
- the equipment used to generate and receive pulse-modulated signal
is generally more complicated and expensive (than for CW
modulation).

126
 Analog Pulse Modulation with Noise (1)
Model for demodulation of analog pulse modulation with noise:

The pulse converter measures the amplitude, duration or position of each

pulse in v(t) and produces:

where µp is modulation constant, εk is measurement error induced by noise.

Assumed that the reconstruction filter has bandwidth B = fs/2, K = Ts = 1/2B
and zero time delay. Hence,

127
 Analog Pulse Modulation with Noise (2)

with

The errors εk are proportional to the sample values of noise n(t) spaced by
Ts. The errors εk are uncorrelated and have zero mean, then

The signal power is: . Hence,

(*)

For PAM signal:

128
 Analog Pulse Modulation with Noise (3)
Therefore, under the best condition with maximum modulation index (µ = 1)
and minimum noise bandwidth (BN = 1/2τ, with τ is pulse duration), we
have:

The average power per modulated pulse is

With pulse rate fs we obtain the received signal power:

Thus, we obtain the final result in form:

This result shows that (S/N)D ≤ γ/2 just like AM CW modulation.

129
 Analog Pulse Modulation with Noise (4)
For PPM and PDM, it offers improvement. If BN ≈ BT, time-position error
variance:

Since the amplitude A is constant, the receive power can be written as

Where τ0 is average pulse duration in PDM or fixed pulse duration in PPM.

Then from (*), we obtain:

For PPM, µ0 = t0, the parameters t0 andτ0 are constrained by

130
 Analog Pulse Modulation with Noise (5)
Then we obtain the upper bound:

For PDM with , we obtain:

131
 Analog Pulse Modulation with Noise (6)
False-pulse threshold effect:
Assume that we can increase BT to increase (S/N)D in PPM and PDM
system. However, the noise power will increase when BT increased. The
false pulses due to noise may occur often. Therefore, the message may be
completely lost (similar to the threshold effect in WBFM).

This effect does not exist in PAM system with synchronization.

To determine the threshold level, assumed that the false pulses sufficiently
infrequent if . For Gaussian noise with
then approximately A ≥ 2σN .

With A2 = SR/τ0 fs we obtain:

This threshold is appreciably less than that of FM. So PPM could be

advantageous for the situations where FM would be below threshold point.

132
133
134
Câu 1: Cho tín hiệu đơn tần m(t)=1,2sin4πt (t:ms) và sóng mang xc(t)=10cos20πt
(t:ms).
a) Tín hiệu m(t) được điều chế biên độ (AM) với chỉ số điều chế μ = 0.5. Vẽ dạng sóng
của tín hiệu sau điều chế. Tìm điều kiện của chỉ số điều chế μ với tín hiệu ngõ vào
m(t) nêu trên sao cho có thể dùng bộ tách sóng đường bao để giải điều chế mà không
bị sái dạng.
b) Tín hiệu m(t) được điều chế hai biên triệt sóng mang (DSB). Tính công suất của tín
hiệu sau điều chế.
c) Thiết kế 1 sơ đồ nguyên lý thực hiện điều chế DSB từ các bộ điều chế AM (chỉ số
điều chế µ, bộ tạo sóng mang có sẵn trong bộ điều chế), bộ khuếch đại và bộ cộng.
d) Tín hiệu m(t) được điều chế đơn biên (LSSB). Viết biểu thức của tín hiệu sau điều
chế.
Câu 2: Cho biết tín hiệu m(t)=0,8sin4πt (t:ms) và sóng mang xc(t)=10cos20πt (t:ms).
a) Tín hiệu m(t) được điều chế tần số (FM) với độ di tần fΔ = 2.5KHz. Vẽ phổ biên độ
(tần số dương) của tín hiệu sau điều chế
b) Tín hiệu m(t) được điều chế pha (PM) với độ di pha φΔ = 100o. Ước lượng băng
thông của tín hiệu sau điều chế.
c) Vẽ 1 sơ đồ nguyên lý của bộ điều chế FM với độ di tần fΔ từ các bộ điều chế PM
(độ di pha φΔ), bộ tạo sóng mang, bộ tích phân/vi phân, bộ khuếch đại và bộ cộng.

135